Refined O3 source analysis method and device based on machine learning, equipment and medium

Abstract

The embodiment of the invention discloses a refined O3 source analysis method and device based on machine learning, equipment and a medium, and belongs to the technical field of environmental sciences, the method comprises the steps that simulation and source analysis are carried out on local O3 through WRF-CMAQ-ISAM, contribution of each pollution source to O3 is obtained, and the contribution of each pollution source to the O3 is obtained; the method comprises the following steps: firstly, optimizing the ozone concentration change, then standardizing meteorological conditions and pollution source contribution by adopting a random sampling-model prediction-concentration averaging machine learning method to obtain ozone concentration change which is not influenced by weather and emission change, and finally, introducing an SHAP attribution method to quantify the influence of meteorological parameters and pollution source change on O3. Through the method, absolute contribution of meteorological parameters and local emission to O3 change can be obtained, so that an important theoretical basis is provided for accurate prevention and control of O3.

Description

Technical Field

[0001] The disclosed embodiments relate to the field of environmental science and technology, and specifically to a refined O3 source analysis method, device, equipment, and medium based on machine learning. Background Art

[0002] The information disclosed in this background technology section is only intended to increase the understanding of the overall background of the present disclosure, and shall not necessarily be regarded as an admission or any form of suggestion that the information constitutes the relevant technology that has been known to ordinary technicians in the field.

[0003] Atmospheric O3 is a secondary pollutant. The precursors of the same emission (volatile organic compounds and nitrogen oxides) show a nonlinear relationship and are also affected by meteorological conditions such as temperature, relative humidity, wind direction and wind speed. This makes the causes of O3 in the atmosphere and its impact on the environment very complex, hindering the prevention and control of O3 pollution. Therefore, it is necessary to construct a method to quantify the actual contribution of meteorological factors from each emission source, strengthen the practicality and scientific nature of management and control measures, and support precise and scientific O3 pollution prevention and control.

[0004] In related technologies, the source analysis of O3 is mainly carried out through the air quality model method, which is based on the concept of "one atmosphere" and takes into account a series of physical and chemical processes after pollutants are emitted into the atmosphere as well as the influence of meteorological factors. Ideally, this method can accurately locate the time and space of pollution sources and quantify the contribution concentration and contribution rate of emission sources, but it cannot quantify meteorological parameters and the contribution of each pollution source under the influence of meteorological conditions, and has poor accuracy. Summary of the invention

[0005] To this end, the embodiments of the present disclosure provide a refined O3 source analysis method, device, equipment, and medium based on machine learning to solve the problem that the source analysis of O3 using physical and chemical principles in related technologies cannot quantify meteorological parameters and the contribution of various pollution sources under the influence of meteorological conditions, resulting in poor data accuracy due to poor accuracy.

[0006] In order to achieve the above objectives, the embodiments of the present disclosure provide the following technical solutions:

[0007] In a first aspect of the embodiments of the present disclosure, a refined O3 source analysis method based on machine learning is provided, comprising the following steps:

[0008] Obtain historical meteorological data and air pollution observation data to determine the air quality model system;

[0009] The O3 concentration is simulated and the O3 source is analyzed through the air quality model system to obtain the contribution of each emission source to O3;

[0010] The O3 source analysis results, meteorological data, and air pollution observation data were used to train the random forest and verify it to obtain a trained random forest.

[0011] The input meteorological conditions and emission contributions were standardized using the “random sampling-model prediction-concentration averaging” method;

[0012] The difference between the standardized ozone dataset and the actual ozone was used as the target variable, and the contribution of each meteorological factor and emission source was used as the input parameter to train the random forest model;

[0013] The previously trained random forest model and each parameter are input into the Shapley addition and interpretation SHAP attribution method, and the contribution of each parameter to O3 is output.

[0014] In one embodiment, the air quality model system includes a mesoscale weather model (WRF), a community multiscale air quality model (CMAQ)-integrated source apportionment method (ISAM).

[0015] In one embodiment, the step of obtaining the contribution of each emission source to O3 comprises:

[0016] The O3 concentration is simulated by the air quality model system to obtain the simulation results;

[0017] Verifying the simulation results;

[0018] ISAM is used to analyze the source of O3 and obtain the contribution of each emission source to O3.

[0019] In one embodiment, a method for simulating O3 concentration includes:

[0020] Based on the FNL reanalysis meteorological data, the mesoscale weather numerical model (WRF) was used to simulate meteorological conditions. According to the meteorological conditions and the local atmospheric pollution source emission inventory data, the local gridded pollution source emission inventory and natural source inventory in the network common data format (netCDF) were constructed through the atmospheric pollutant emission source inventory processing model (SMOKE) and the natural source emission inventory processing model (MEGAN), and the O3 concentration was simulated through the community multiscale air quality model (CMAQ).

[0021] When the simulation is good, the ISAM module in CMAQ is used to analyze the source of O3 and obtain the contribution of each emission source to O3.

[0022] In one embodiment, after the above process is completed, the simulation effect of the O3 concentration is verified, and if the simulation is poor, the list and parameters are adjusted;

[0023] The parameters of the simulation verification of O3 include: mean deviation (MB), standardized deviation (NMB), standardized mean error (ME), root mean square error (RMSE), correlation coefficient (R), and consistency index (IOA). A good simulation requires NMB to be less than 15%, R to be greater than 0.75, and IOA to be greater than 0.75.

[0024] In one embodiment, the training of the random forest includes

[0025] In the Python environment, we used the obtained source analysis results, meteorological data, and observation data to train the random forest, and adjusted the parameters for our specific application. The results required that the R of the O3 predicted value of the trained random forest model was 1.0447 W / m. 2 Above 0.90.

[0026] In one embodiment, input meteorological conditions and emission contributions are normalized, including:

[0027] Build the above random forest model that can predict ozone concentration;

[0028] Keep the Unix time column unchanged, and randomly sample from the complete historical meteorological data based on each data node to form a new data set with a disrupted order;

[0029] The sampling results are input into the random forest model to predict the ozone concentration.

[0030] The above steps were repeated 1000 times, and the prediction results were averaged to obtain the ozone concentration under almost all meteorological and emission conditions, that is, the ozone result with standardized input parameters.

[0031] In one embodiment, the trained random forest model and the contribution of each meteorological and emission source are input into the SHAP method, and hyperparameter tuning is performed through random search method and grid search method to return the hyperparameter combination with the best simulation effect, and finally output the contribution of each input parameter to O3.

[0032] In a second aspect of the embodiments of the present disclosure, a refined O3 source analysis device based on machine learning is provided, which is used to implement any of the above methods, including:

[0033] A data acquisition module configured to acquire historical meteorological data and air pollution observation data;

[0034] An air quality simulation module, configured to simulate and source-apportion O3 concentrations through an air quality model;

[0035] The machine learning prediction actual O3 module is configured to train the random forest using the acquired source apportionment results, meteorological data, and observation data, and to perform validation to obtain a trained random forest;

[0036] The normalization module is configured to normalize the input meteorological conditions and emission contributions using the “random sampling-model prediction-concentration averaging” method;

[0037] The machine learning module for predicting O3 changes is configured to train a random forest model using the standardized ozone dataset and the actual ozone difference as target variables and the contribution of each meteorological factor and emission source as input parameters;

[0038] The attribution module is configured to input the previously trained random forest model and each parameter into the SHAP attribution method, and output the contribution of each parameter to O3.

[0039] In a third aspect of the embodiments of the present disclosure, there is provided an electronic device, comprising an input device and an output device, and further comprising

[0040] a processor adapted to implement one or more instructions; and

[0041] A computer-readable storage medium having stored thereon at least one instruction, at least one program, a code set or an instruction set, wherein the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the steps of any of the above methods.

[0042] In a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, characterized in that at least one instruction, at least one program, a code set or an instruction set is stored on the computer-readable storage medium, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the steps of any of the above methods.

[0043] According to the embodiments of the present disclosure, the method has the following advantages: local O3 is simulated and source parsed by WRF-CMAQ-ISAM to obtain the contribution of each pollution source to O3, and then the machine learning method of "random sampling-model prediction-concentration average" is used to standardize the meteorological conditions and the contribution of pollution sources to obtain the change of ozone concentration that is not affected by meteorological and emission changes, and finally the SHAP attribution method is introduced to quantify the impact of meteorological parameters and pollution source changes on O3. Through the present invention, the absolute contribution of meteorological parameters and local emissions to O3 changes can be obtained, thereby providing an important theoretical basis for the precise prevention and control of O3. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the related technologies, the following briefly introduces the drawings required for the embodiments or the related technical descriptions. Obviously, the drawings described below are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0045] Figure 1 is a flow chart of a refined O3 source analysis method based on machine learning according to an exemplary embodiment;

[0046] Figure 2 is a structural block diagram of a refined O3 source analysis device based on machine learning according to an exemplary embodiment;

[0047] Figure 3 The present invention is a flow chart of a refined O3 source analysis method based on machine learning according to an exemplary embodiment;

[0048] Figure 4 A schematic diagram of a standardized technical principle in a method according to an exemplary embodiment;

[0049] Figure 5 FIG. 4 is a flow chart of the SHAP attribution method in a method according to an exemplary embodiment. DETAILED DESCRIPTION

[0050] The following is a description of the implementation methods of the present disclosure by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present disclosure, but not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present disclosure.

[0051] The above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform, or by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the relevant technology, can be embodied in the form of a software product, and cannot be associated with the limitation of the implementation of the method of the present disclosure due to the limitation of the hardware structure platform. Therefore, the embodiments of the present method are all applicable to the electronic device and storage medium, and can achieve the same or similar beneficial effects.

[0052] like Figure 1 As shown, it shows a flow chart of a refined O3 source analysis method based on machine learning provided by an embodiment of the present invention. The method comprises the following steps:

[0053] Obtain historical meteorological data and air pollution observation data to determine the air quality model system;

[0054] The O3 concentration is simulated and the O3 source is analyzed through the air quality model system to obtain the contribution of each emission source to O3;

[0055] The O3 source analysis results, meteorological data, and air pollution observation data were used to train the random forest and verify it to obtain a trained random forest.

[0056] The input meteorological conditions and emission contributions were standardized using the “random sampling-model prediction-concentration averaging” method;

[0057] The difference between the standardized ozone dataset and the actual ozone was used as the target variable, and the contribution of each meteorological factor and emission source was used as the input parameter to train the random forest model;

[0058] The previously trained random forest model and each parameter are input into the Shapley addition and interpretation SHAP attribution method, and the contribution of each parameter to O3 is output.

[0059] like Figure 2 As shown in the figure, the implementation of the method is mainly divided into two parts, namely the air quality model part and the machine learning part. The main purpose of the air quality model part is to obtain the contribution of each emission source to ozone. Considering that O3 is a secondary pollutant without primary emission, the emission intensity and category of precursors from different emission sources are different, so the contribution of emission sources to O3 obtained by the air quality model is used to comprehensively represent it.

[0060] In one embodiment, the air quality model system includes the mesoscale weather model WRF-the community multiscale air quality model CMAQ-the integrated source analysis method ISAM.

[0061] In one embodiment, the step of obtaining the contribution of each emission source to O3 includes:

[0062] The O3 concentration is simulated by the air quality model system to obtain the simulation results;

[0063] Verify the simulation results;

[0064] The source of O3 was analyzed using the ISAM method to obtain the contribution of each emission source to O3.

[0065] In one embodiment, a method for simulating O3 concentration includes:

[0066] Based on the FNL reanalysis meteorological data, the mesoscale weather numerical model WRF was used to simulate meteorological conditions. According to the meteorological conditions and the local atmospheric pollution source emission inventory data, the atmospheric pollutant emission source inventory processing model SMOKE and the natural source emission inventory processing model MEGAN were used to construct the local gridded pollution source emission inventory and natural source inventory in the network common data format (netCDF), and the O3 concentration was simulated through the community multi-scale air quality model CMAQ.

[0067] When the simulation is good, the ISAM module in CMAQ is used to analyze the source of O3 and obtain the contribution of each emission source to O3.

[0068] In one embodiment, after the above process is completed, the simulation effect of the O3 concentration is verified, and if the simulation is poor, the list and parameters are adjusted;

[0069] The parameters of the simulation verification of O3 include: mean deviation MB, standardized deviation NMB, standardized mean error ME, root mean square error RMSE, correlation coefficient R, and consistency index IOA. A good simulation requires NMB to be less than 15%, R to be greater than 0.75, and IOA to be greater than 0.75.

[0070] The second part of machine learning mainly includes two major sections, namely standardization and SHAP attribution, such as Figure 2 and Figure 3 shown.

[0071] The main steps of standardization include: In one embodiment, the training of random forest includes

[0072] In the Python environment, we used the obtained source analysis results, meteorological data, and observation data to train the random forest, and adjusted the parameters for our specific application. The results required that the R of the O3 predicted value of the trained random forest model was 1.0447 W / m. 2 Above 0.90.

[0073] In one embodiment, input meteorological conditions and emission contributions are normalized, including:

[0074] Build the above random forest model that can predict ozone concentration;

[0075] Keep the Unix time column unchanged, and randomly sample from the complete historical meteorological data based on each data node to form a new data set with a disrupted order;

[0076] The sampling results are input into the random forest model to predict the ozone concentration.

[0077] The above steps were repeated 1000 times, and the prediction results were averaged to obtain the ozone concentration under almost all meteorological and emission conditions, that is, the ozone result with standardized input parameters.

[0078] The SHAP method is a machine learning explanation model based on cooperative game theory, which can quantify the impact of each feature on a single prediction result. The core principles of the SHAP method are as follows:

[0079]

[0080] Where f is the trained machine learning model, φ0(f) is the baseline value of the model, that is, the average value of the model prediction results, and M is the number of features of the model. i (f, x) is the SHAP value of feature i, that is, the contribution of feature i to the prediction result f(x). i (f,x)>0(φ i (f, x) < 0), it means that feature i will increase (decrease) f(x), while φ i The absolute value of (f, x) indicates the importance of feature i.

[0081]

[0082] in, is the set of all feature sorts, Is in the collection The set of all features that precede feature i in .

[0083] The application of the SHAP attribution method mainly includes the following steps:

[0084] Step 1: Subtract the O3 dataset output from the standardization step from the actual ozone dataset to obtain the meteorological and emission-variable ozone dataset.

[0085] Step 2: Using the data set output from step 1 as the target variable and the contribution of each meteorological factor and emission source as input parameters, the random forest model is trained and validated.

[0086] Step 3: The random forest model trained in step 2 and the meteorological and meteorological emission datasets are input into the SHAP method, and the output result is the contribution of each input parameter to the change of O3.

[0087] In one embodiment, the trained random forest model and the contribution of each meteorological and emission source are input into the SHAP method, and hyperparameter tuning is performed through random search method and grid search method to return the hyperparameter combination with the best simulation effect, and finally output the contribution of each input parameter to O3.

[0088] In a second aspect of the embodiments of the present disclosure, a refined O3 source analysis device based on machine learning is provided, which is used to implement any of the above methods, such as Figure 2 Shown, including

[0089] A data acquisition module configured to acquire historical meteorological data and air pollution observation data;

[0090] An air quality simulation module, configured to simulate and source-apportion O3 concentrations through an air quality model;

[0091] The machine learning prediction actual O3 module is configured to train the random forest using the acquired source apportionment results, meteorological data, and observation data, and to perform validation to obtain a trained random forest;

[0092] The normalization module is configured to normalize the input meteorological conditions and emission contributions using the “random sampling-model prediction-concentration averaging” method;

[0093] The machine learning module for predicting O3 changes is configured to train a random forest model using the standardized ozone dataset and the actual ozone difference as target variables and the contribution of each meteorological factor and emission source as input parameters;

[0094] The attribution module is configured to input the previously trained random forest model and each parameter into the SHAP attribution method, and output the contribution of each parameter to O3.

[0095] In a third aspect of the embodiments of the present disclosure, there is provided an electronic device, comprising an input device and an output device, and further comprising

[0096] a processor adapted to implement one or more instructions; and

[0097] A computer-readable storage medium having stored thereon at least one instruction, at least one program, code set or instruction set, wherein the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the steps of any of the above methods.

[0098] In a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, characterized in that at least one instruction, at least one program, code set or instruction set is stored on the computer-readable storage medium, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the steps of any of the above methods.

[0099] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0100] Although the present disclosure has been described in detail above with general descriptions and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements may be made to the present disclosure. Therefore, these modifications or improvements made without departing from the spirit of the present disclosure are within the scope of protection claimed by the present disclosure.

Claims

1. A refined O3 source apportionment method based on machine learning, characterized in that: The following steps are included Obtain historical meteorological data and air pollution observation data to determine the air quality model system; The O3 concentration is simulated and the O3 source is analyzed through the air quality model system to obtain the contribution of each emission source to O3; The O3 source analysis results, meteorological data, and air pollution observation data were used to train the random forest and verify it to obtain a trained random forest. The input meteorological conditions and emission contributions were standardized using the "random sampling-model prediction-concentration averaging" method; The difference between the standardized ozone dataset and the actual ozone was used as the target variable, and the contribution of each meteorological factor and emission source was used as the input parameter to train the random forest model; The previously trained random forest model and each parameter are input into the Shapley addition and interpretation SHAP attribution method, and the contribution of each parameter to O3 is output.

2. The refined O3 source analysis method based on machine learning according to claim 1 is characterized in that: The air quality model system includes a mesoscale weather model, a community multi-scale air quality model, and a comprehensive source analysis method.

3. The refined O3 source analysis method based on machine learning according to claim 2 is characterized in that: The step of obtaining the contribution of each emission source to O3 comprises: The O3 concentration is simulated by the air quality model system to obtain the simulation results; Verifying the simulation results; The comprehensive source analysis method is used to analyze the sources of O3 and obtain the contribution of each emission source to O3.

4. The refined O3 source analysis method based on machine learning according to claim 3 is characterized in that: The method for simulating O3 concentration includes: Based on the FNL reanalysis meteorological data, a mesoscale weather numerical model is used to simulate meteorological conditions. According to the meteorological conditions and local atmospheric pollution source emission inventory data, the atmospheric pollutant emission source inventory processing model and the natural source emission inventory processing model are used to construct local gridded pollution source emission inventories and natural source inventories in a network common data format. The O3 concentration is simulated through a community multi-scale air quality model. When the simulation is good, the ISAM module in CMAQ is used to analyze the source of O3 and obtain the contribution of each emission source to O3.

5. The refined O3 source analysis method based on machine learning according to claim 4 is characterized in that: After the above process is completed, the simulation effect of O3 concentration is verified. If the simulation is poor, the list and parameters are adjusted; The parameters of O3 simulation verification include: mean deviation, standardized deviation, standardized mean error, root mean square error, correlation coefficient, and consistency index. A good simulation requires that the standardized deviation is less than 15%, the correlation coefficient is greater than 0.75, and the consistency index is greater than 0.

75.

6. The refined O3 source apportionment method based on machine learning according to claim 1, characterized in that: The training of random forest includes In the Python environment, we used the obtained source analysis results, meteorological data, and observation data to train the random forest, and adjusted the parameters for our specific application. The results required that the R of the O3 predicted value of the trained random forest model was 1.0447 W / m. 2 Above 0.

90.

7. The refined O3 source apportionment method based on machine learning according to claim 6, characterized in that: Normalize input meteorological conditions and emission contributions, including: Build the above random forest model that can predict ozone concentration; Keep the Unix time column unchanged, and randomly sample from the complete historical meteorological data based on each data node to form a new data set with a disrupted order; The sampling results are input into the random forest model to predict the ozone concentration. The above steps were repeated 1000 times, and the prediction results were averaged to obtain the ozone concentration under almost all meteorological and emission conditions, that is, the ozone result with standardized input parameters.

8. The refined O3 source apportionment method based on machine learning according to claim 7 is characterized in that: The trained random forest model and the contribution of each meteorological and emission source are input into the SHAP method, and hyperparameter tuning is performed through random search method and grid search method. The hyperparameter combination with the best simulation effect is returned, and finally the contribution of each input parameter to O3 is output.

9. A refined O3 source analysis device based on machine learning, used to implement the method according to any one of claims 1 to 8, characterized in that: include A data acquisition module configured to acquire historical meteorological data and air pollution observation data; An air quality simulation module, configured to simulate and source apportion O3 concentrations through an air quality model; The machine learning prediction actual O3 module is configured to train the random forest using the acquired source apportionment results, meteorological data, and observation data, and to perform validation to obtain a trained random forest; The normalization module is configured to normalize the input meteorological conditions and emission contributions using the "random sampling-model prediction-concentration averaging" method; The machine learning module for predicting O3 changes is configured to train a random forest model using the standardized ozone dataset and the actual ozone difference as target variables and the contribution of each meteorological factor and emission source as input parameters; The attribution module is configured to input the previously trained random forest model and each parameter into the SHAP attribution method, and output the contribution of each parameter to O3.

10. An electronic device, comprising an input device and an output device, characterized in that: Also included is a processor adapted to implement the one or more instructions; and A computer-readable storage medium having stored thereon at least one instruction, at least one program, a code set or an instruction set, wherein the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the steps of any method in claims 1-7.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the steps of any method in claims 1-7.

Images